APPARATUS AND METHOD FOR PERFORMING ARTIFICIAL INTELLIGENCE (Al) ENCODING AND Al DECODING ON IMAGE

2. The electronic device of claim 1, wherein the NN setting information includes weights and biases of filter kernels in at least one convolution layer of the up-scaling NN.

4. The electronic device of claim 3, wherein the first loss information is generated based on a result of comparing a quality parameter of the second training image with a quality parameter of the at least one original image.

5. The electronic device of claim 4, wherein the third loss information is generated based on a result of comparing a feature-related parameter of the second training image with a feature-related parameter of the at least one original image.

6. The electronic device of claim 4, wherein the second loss information is related to a spatial complexity of the first training image.

8. The electronic device of claim 1, wherein the NN setting information updated for performing the AI up-scaling further includes weight residual information and bias residual information indicating a difference between a weight and a bias of all or some of filter kernels in all or some of convolution layers in the up-scaling NN before the weight and the bias are updated and a weight and a bias of the all or some of the filter kernels in the all or some of the convolution layers in the up-scaling NN after the weight and the bias are updated when the flag information indicates to perform the AI up-scaling by using a filter kernel of a convolution layer in the updated NN.

9. The electronic device of claim 8, wherein the weight residual information and the bias residual information are information encoded using one of differential pulse code modulation (DPCM), run-length coding (RLC), and Huffman coding schemes.

10. The electronic device of claim 8, wherein the weight residual information and the bias residual information are information about a weight residual and a bias residual generated via model compression.

11. The electronic device of claim 10, wherein the model compression comprises at least one of pruning or quantization.

12. The electronic device of claim 1, wherein the NN setting information updated for performing the AI up-scaling further includes information about a weight residual and a bias residual obtained by performing frequency transformation when the flag information indicates to perform the AI up-scaling by using a filter kernel of a convolution layer in the updated NN, the information about the weight residual and the bias residual indicating a difference between a weight and a bias of all or some of filter kernels in all or some of convolution layers in the up-scaling NN before the weight and the bias are updated and a weight and a bias of the all or some of the filter kernels in the all or some of the convolution layers in the up-scaling NN after the weight and the bias are updated.

13. The electronic device of claim 1, wherein the NN setting information updated for performing the AI up-scaling further includes information obtained by entropy-encoding a weight and a bias of all or some of filter kernels in all or some of convolution layers in the up-scaling NN after the weight and the bias are updated, based on context model information regarding a weight and a bias of the all or some of the filter kernels in the all or some of the convolution layers in the up-scaling NN before the weight and the bias are updated when the flag information indicates to perform the AI up-scaling by using a filter kernel of a convolution layer in the updated NN.

14. The electronic device of claim 1, wherein the NN setting information includes NN information updated for performing the AI up-scaling of the at least one second image corresponding to the at least one original image via joint training of the up-scaling NN and the down-scaling NN used for the AI down-scaling of the at least one original image.

16. A non-transitory computer-readable recording medium having recorded thereon a program for executing the method of claim 15.